Machines with Memory

1 / 36

# Machines with Memory - PowerPoint PPT Presentation

Machines with Memory. Chapter 3 (Part A). Overview. Though every finite computational task can be realized by a circuit, it is not practical – we cannot afford to design a special circuit for each one

I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.

## PowerPoint Slideshow about 'Machines with Memory' - mora

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

### Machines with Memory

Chapter 3 (Part A)

Overview
• Though every finite computational task can be realized by a circuit, it is not practical – we cannot afford to design a special circuit for each one
• Machines with memory allow for reuse of equivalent circuits to realize functions of high circuit complexity
• Discuss the following machines
• Finite State Machines (FST)
• Random Access Machines (RAM)
• Turing Machines (TM)
Overview (cont.)
• Show via simulation that RAMs and TMs are universal machines
• RAMs can simulate any FSM
• RAMs and TMs can simulate each other
• RAMs and TMs are therefore excellent reference models of computation
• Simulate with circuits computations performed by FSMs, RAMs, and TMs
• Establish that all computations are constrained by available time and space resources
• Show there exists an O(log ST)-space, O(ST)-time program to write descriptions of circuits simulating FSMs, RAMs, and TMs
• Provide first examples of P-complete and NP-complete problems
Output

Input

,

Memory

Finite-StateMachines
Finite-StateMachines (cont.)
• A FSM can be represented with a transition diagram:
• This FSM enters the final state q1 when an odd number of 1’s are read at input
• A realization of the EXCLUSIVE OR function on an arbitrary number of inputs
Finite-StateMachines (cont.)
• This variant is a Moore machine – output associated with each state
• Another variant, called the Mealey machine, output is generated with each state transition
• Can be shown that one can simulate the other and vice-versa
(T)

fM (q(0) , w1,…, wT) = (q(T) , y1 , y2 , … , yT )

Functions Computedby FSMs
• Allow our FSMs only to compute functions that receive inputs and produce outputs at data-independent times
• Trace computation starting at initial state

q(0) = s reads input symbol w1 enters state

q(1) = ( q(0), w1) producing output y1 reads input symbol w2 enters state

q(2) = ( q(1), w2)producing output y2 reads input symbol w3 enters state :

q(T) = ( q(T1), wT)producing output yT

• Function thus computed can be expressed as
(T)

fM (q(0) , w1,…, wT) = (q(T) , y1 , y2 , … , yT )

Functions Computedby FSMs (cont.)
• “General” function computed by an FSM in T steps:
• Memory serves to “remember” the state where computation currently is:
• “the role of memory is to hold intermediate results on which the logical circuitry of the machine can operate in successive cycles”
Most FSMs used in T-step computations compute only subfunctions of this general function

(T)

fM

Functions Computedby FSMs (cont.)
• Example: FSM that forms the EXCLUSIVE OR of n variables
• Only the final output symbol is used (all intermediate “output” symbols are ignored)
Computational tasks are modeled by binary functions, for example, the function computed by a general FSM if all components involved in the definition (, , Q, , , s, F) are all assumed to be encoded in binary.

(T)

(T)

fM

fM

Functions Computedby FSMs (cont.)
• Use this general function in deriving space-time product inequalities for RAMs, in establishing a connection between Turing time and circuit complexity, and in the definition of certain P-complete and NP-complete problems
• First result involves time and circuit size and depth of a logical circuit realizing the function computed by an FSM:
Computational Inequalitiesfor the FSM
• Result follows from definitions of C()and D(), and previous result (Lemma 2.4.1) about circuit size and depth of subfunctions vis-à-vis function of reference
• Define TC(,) for an FSM M as the equivalent number of logic operations performed by M in T steps
Computational Inequalitiesfor the FSM (cont.)
• First inequality above says that “the number of equivalent logic operations performed byan FSM to compute a function f must be at least the minimum number of gates necessary to compose f with a circuit”
• Place upper limits on the size and depth complexities of functions computable in a bounded number of steps by an FSM
THEOREM 3.1.2 Every subfunction of the function , computable by an FSM on n inputs, is computable by a Boolean circuit and vice versa.

(T)

fM

Circuits are Universal for Bounded FSM Computations
• Next result identifies logical circuits with bounded-time FSMs
• One direction () is implied by previous discussion (cf. Fig. 3.3)
THEOREM 3.1.2 Every subfunction of the function , computable by an FSM on n inputs, is computable by a Boolean circuit and vice versa.

Start

s/y1

(T)

1

0

fM

q00/y10

q00/y10

1

0

1

0

:

:

:

:

Circuits are Universal for Bounded FSM Computations (cont.)
• Other direction () is established by considering a “binary tree” transition diagram describing an FSM with the start state as root and branching into two states at each node to represent transition in response to two possible input values. If there are n inputs, then the transition diagram would have n levels and at most 2n+11states
THEOREM 3.1.2 Every subfunction of the function , computable by an FSM on n inputs, is computable by a Boolean circuit and vice versa.

(T)

fM

Circuits are Universal for Bounded FSM Computations (cont.)
• Example:
Interconnections of Finite-State Machines
• Synchronous FSMs read inputs, advance from state to state, and produce outputs in synchrony
• Can achieve greater computational power and flexibility by interconnecting synchronous FSMs, for example
Interconnections of Finite-State Machines (cont.)
• Analogous inequality results for interconnected FSMs:
• Proof is immediate from the definition of synchronous FSMs and previous inequality results
Start

qi

qj

qk

s

f

Nondeterministic Finite-State Machines
• Potentially more general FSM model obtained from “normal” FSMs (which are deterministic) by allowing state transition relations rather than strict functions
• The equivalent of being able to “guess” the correct state transition in order to arrive eventually (if at all possible) at an accepting state – nondeterministic FSM or NFSM
Nondeterministic Finite-State Machines (cont.)
• NFSMs can be thought of as deterministic FSMs with additional input – choice input – in addition to standard input
• Choice agent supplies the choice input and disambiguates state transition whenever nondeterminism is involved
0,1

Start

1

1

f /1

q1/0

q0/0

1

1

f/1

1

Start

q0/0

0

1

q1/0

0

Nondeterministic Finite-State Machines (cont.)
• NFSM that accepts strings that are binary representations of odd integers (e.g., 1, 11, 101, 111, 1001, …):
• Equivalent DFSM:
P=NP

vs.

P

NP

Nondeterministic Finite-State Machines (cont.)
• NFSMs and DFSMs recognize the same class of languages – non-determinism does not make an FSM more powerful
• For Turing Machines (TMs), the difference between the power of deterministic vs. non-deterministic is not known – at the heart of the PNP issue
Designing Sequential Circuits
• Sequential circuits – circuits constructed from logic circuits and clocked (binary) memory devices
• Inputs to logic circuit may be external input and/or outputs from binary memory units
• Example has one EXCLUSIVE OR gate and one clocked memory unit which feeds its output back to the gate
• One external input provides other input to the EXCLUSIVE OR gate
• Sketch a procedure to design sequential circuits that realize any FSM – show relationship between FSMs and logic circuits
Designing Sequential Circuits (cont.)
• First step: assign unique binary tuples to each FSM input symbol, output symbol, and state
• Second step: tables for the next-state function  and output function  are produced based on FSM description and binary encoding of first step
• Third step: logical circuits are designed to realize the binary functions of second step
• Fourth step: logical circuits are connected to clocked binary memory devices to produce the desired sequential circuit realizing the FSM
Designing Sequential Circuits (cont.)
• Example: two-bit adder (cf. Fig. 2.14)
• First step: assign unique binary tuples to input symbols, output symbols, and states
Designing Sequential Circuits (cont.)
• Example: two-bit adder (cf. Fig. 2.14)
• Second step: tables for the next-state function  and output function  are produced

(q0,00)=q0

(00,00)=00

Designing Sequential Circuits (cont.)
• Third & fourth steps: build logical and sequential circuits
Random Access Machines
• RAMs – model essential features of traditional serial computers
• Composed of two synchronous interconnected FSMs
• A CPU that has a modest number of local registers (memory device); and
• A random-access memory that has a large number of local registers
RAM Architecture
• CPU implements a fetch-and-execute cycle, alternatively
• Retrieving an instruction from memory (stored-program concept, cf. von Neumann); and
• Executing the retrieved instruction
• Instructions are of the following five categories:
• Arithmetic/logical instructions
• Jump instructions for breaking normal sequential “flow of logic”
• I/O instructions
• Halt instruction
• Random-Access Memory allows for equal-time access to any of its component registers
• One output word (out_wrd)
• Three input words: an address (addr), a data word (in_wrd), and a command (cmd)
The RAM as FSM
• The CPU is an FSM –
• receives input from the random-access-memory and external sources
• sends output to random-access-memory and output port
• its state is a function of the content of its registers
The RAM as FSM (cont.)
• The random-access memory is an FSM –
• receives input from the CPU
• sends output to the CPU
• its state is a function of the content of its registers
RAM Programs
• Expressed as assembly-language programs to make them more readable (use of mnemonics and labels instead of bit patterns
RAM Programs (cont.)
• Sample assembly-language programs, one for a simple addition, and the other for squaring a given value
• Second program makes use of first program as a “subroutine”
Universality of the RAM
• Define the notion of “universality”
• Universality theorem for RAMs vs. FSMs
• Proof sketch:
• FSM is completely defined by its next-state and output functions  just need to write a fixed-length RAM program implementing both functions and recording the output and state values in the RAM memory
• RAM program is run repeatedly to simulate FSM function
• RAM space required is fixed: storage for RAM simulation programs for  and , plus register to store FSM state
Universality of the RAM (cont.)
• RAMs can simulate FSMs
• Since RAM components (CPU and bounded random-access-memory) are themselves FSMs, a RAM can simulate any other RAM
• Another “universality” result: RAMs can execute RAM programs
• Two “flavors” of RAM to execute RAM programs:
• RAM program is stored in registers specially allocated to the RAM program (loaded onto CPU)
• RAM program is stored in registers of the random-access-memory (RASP model)
• For later discussion (if time permits)
Computational Inequalitiesfor the RAM
• Use prior inequalities for FSMs since bounded-memory RAM components are themselves FSMs
• Use results from Sec. 3.10.6 and Sec. 3.5 (not discussed) to obtain the following results:
• The quantity ST is again pivotal here – another incarnation of the “space-time” tradeoff